A driver’s licence for algorithms

PhD researcher Nassim Motamedidehkordi (left) talking to a PTV employee

Estimated reading time: 6 minutes

Everybody is talking about connected automated vehicles. But what do they have to do with machine learning and microsimulation? We talked to Nassim Motamedidehkordi, a PhD researcher at the Technical University in Munich, Germany, about investigating driver behaviour, the reason why she studied intelligent transportation systems and that a driver’s licence is not only something for humans.

How does machine learning help model and simulate the behaviour of CAVs?
Let’s say we want to model conventional cars, most of the already available simulation software are based on rule‑based models. They are perfect for many applications, however in order to model the reality and variations among the human drivers and have credible results, the models should be carefully calibrated and validated for each specific situation. Nowadays that we have better technology to collect and store data in databases and cheaper computing power, especially in parallel computing, we can also use statistical tools, like machine learning, to learn the driving behaviour from this vast amount of data.

As you know, the focus of my PhD thesis project is modelling the driving behaviour with machine learning tools. The main reason that I decided to use machine learning in modelling, specifically the behaviour of automated vehicles, is the major challenge of mixed traffic situations. In the beginning phase, we will not have a high penetration rate of AVs, and in this context we have to make sure that we avoid too conservative or too aggressive driver behaviour of AVs. The driver behaviour should be socially acceptable, although it can be that in different driving cultures, this social acceptance has different meanings. That motivated me to try and learn driver behaviour from the traffic observation data.

When researching the behaviour of CAVs, what are the different challenges looking at scenarios on motorways and in inner cities?
In urban areas, the complexity and diversity of the environments in which the CAV operates increase and the vehicle faces more traffic users, including pedestrians and cyclists. Dealing with the traffic situations with these road users is a huge challenge for the developers and researchers of automated vehicles. These road users have a lot freedom in their movement and predicting their behaviour is way more challenging than predicting the behaviour of a vehicle driving in one direction on the motorway.

The other challenge is the limitation of sensors which is more obvious in urban areas. The urban infrastructure can, for example, easily block the field of view of sensors. Let’s say you have a building that doesn’t let the sensors of an automated vehicle see around the corner at an intersection – a situation you don’t have on a motorway. Some people believe adding a communication component can tackle the sensors’ limitation and get some backup support from the communication. Even if the sensor cannot see directly what is happening just around the corner, the communication might help the vehicle to avoid the safety-critical and deadlock situations.

How crucial is vehicle-to-infrastructure communication in urban centres?
The communication is a component which would be an add-on that has the potential to increase the traffic efficiency and safety. Some car manufacturers aim to drive automated without relying on any specific communication. In that case, we just have to wait and see how long they need to make the technology mature and safe enough. But dealing with deadlock and critical situations in urban areas without the communication, it will be really challenging, I believe.

How can microsimulation help here and to what extend can it replace field tests of CAVs?
Microsimulation tools have been applied in ITS and automated vehicles studies as a cheap alternative to field tests for many years. All the trial-and-errors can be done in the microsimulation and the vehicles can be tested in the field, on the road, when it is safe. The main benefits of testing in the simulation is that the traffic scenarios are reproducible and the dangerous situations can be tested. Another benefit of microsimulation, or of simulation tools in general, is that they provide the opportunity to project the impact of automated vehicles in the future. In the initial phase when we don’t have them on our streets or a very low penetration rate of them in the fleet, a lot of tests and developments can be done in simulation tools and we can see, what could happen in the future, in thirty, forty or fifty years.

The transportation sector is undergoing drastic changes at the moment with CAVs just being one aspect among many. What made you focus on that topic for your PhD and why did you choose an engineering program when you entered university?
I studied civil engineering for my Bachelor’s and I originally come from Teheran, a city with a population of around 8.4 million people. The whole metropolitan area has 15 million inhabitants. Traffic congestion and air pollution are the two main problems the city is facing. Back then, when I used to live in Teheran, I was so annoyed by the fact that I wasted two hours of my time every single day behind the wheel just to go to university. That was my personal motivation behind it and I was also always enthusiastic about emerging technologies. That is why I chose intelligent transportation systems for my Master’s – to learn how you can optimise transportation systems with these new technologies and I decided to do my PhD also in the same field.

You talked about the dense traffic in Teheran, does that mean you don’t like driving anymore? What is you favourite mode of transport?
For short distances I like cycling. But if I have to commute to work or university, I would rather use public transport because in these thirty minutes of commuting I can read a book or answer emails, that’s not possible when you’re behind the wheel – at least not now until we’ll have fully autonomous driving.

What would you tell people who are sceptical about CAVs, especially about the intermediate phase of conventional vehicles with human drivers sharing the same road with CAVs?
At the moment, there is still a long way to go and we have to make sure that CAVs are safe enough before we have a high penetration rate of them on our streets. But on the other hand, I know, that especially when artificial intelligence and machine learning come into play, some people say, “No, we can’t trust these algorithms”. I don’t believe that these new methods are going to solve all of our problems, but they are tools which can help us. A good example is the collaboration between the German Technical Inspection Association (TÜV Süd) and the German Center for Artificial Intelligence: They started a joint project about a certificate for driving behaviour algorithms, similar to a driver’s licence. When you turn 18 and want to get your driver’s licence, you take driving lessons until you pass the test. This could also be the case for automated vehicles. Of course, the algorithms still have to learn and improve first, but at some point they could then do the licencing and get their driver’s licence.

Thank you, Nassim, for your time.


Are you looking for a microsimulation software that provides a robust set of functionalities for conducting virtual reality tests? Click here and try PTV Vissim 30 days for free.